2018 PAA Short Course on Bayesian Small Area Estimation using Complex Data Introduction and Overview

Transcription

1 2018 PAA Short Course on Bayesian Small Area Estimation using Complex Data Introduction and Overview Sam Clark 1, Zehang Richard Li 2, Jon Wakefield 2,3 1 Department of Sociology, Ohio State University, 2 Departments of Statistics, University of Washington, 3 Departments of Biostatistics, University of Washington 1 / 6

2 Introductions Sam is a demographer with interest in population health in Africa Mortality and its determinants Epi-demographic transitions Experience with demographic surveillance system data collection Interests in population indicator measurement Richard is a statistician, completing his thesis at UW, with interests in Verbal autopsy Bayesian methods and computation Estimation of subnational variation in U5MR Has lead the computational aspects of the U5MR project, including the creation of the SUMMER package Jon is a statistician with longstanding interests in Bayesian statistics Geospatial models and applications in spatial epidemiology Survey sampling and design effects Small-area estimation All three work with IGME group on estimating subnational variation in U5MR. 2 / 6

3 Logistics Demonstrations of methods via R implementations will be carried out in class. Students are encouraged to follow along. Code and other materials (course notes, papers) are available at the course website: 3 / 6

4 Overview Objective Data Global and Spatial Smoothing Bayesian Modeling Survey Sampling Implementation Mapping Totals and averages for a quantity of interest, by area Surveys with a complex design. If small or no samples in some areas, there is high instability To reduce instability, use the totality of data to smooth both locally and globally over space Is convenient/designed for smoothing Required to describe and analyze the sample In R programming environment, with survey and INLA packages. Maps of uncertainty, accompanied with uncertainty, GIS Lecture 1 5, 6 3, 4 Lecture 1 5, 6, 7 Lecture 2 2, 8 4 / 6

5 Course Outline DAY 1: Lecture 1: Bayesian Statistics (Wakefield). Motivation; Bayesian learning, Probability and Bayes theorem; Standard distributions and conjugacy (binomial and normal distributions in detail) Coffee Break Lecture 2: Introduction to R (Li): Introduction to R and RStudio. Examples of normal and binomial sampling, introduction to GIS in R Lunch Break Lecture 3A: Hierarchical Bayes Modeling (Wakefield). Motivation; Non-spatial hierarchical models for normal data; Non-spatial hierarchical models for binomial data Lecture 3B: Hierarchical Bayes Modeling in R (Li). R component: estimation and mapping for hierarchical Bayes models. Introduction to INLA. Simple SAE Coffee Break Lecture 4A: Hierarchical Spatial Bayes Modeling (Wakefield). Spatial hierarchical models for normal data; Spatial hierarchical models for binomial data; Overview of spatial random effects models; normal and binomial examples Lecture 4B: Hierarchical Spatial Bayes Modeling in R (Li). R component: Discrete spatial modeling with INLA. 5 / 6

6 Course Outline DAY 2: Lecture 5A: Survey Sampling (Wakefield). Overview; Simple random sampling; Stratified simple random sampling; Cluster sampling Lecture 5B: Survey Sampling in R (Li): Survey sampling in R, the survey package Coffee Break Lecture 6A Introduction to SAE: (Wakefield): Overview of SAE models; Multistage sampling; Simple SAE Models Lecture 6B: Introduction to SAE in R (Li): Simple SAE in R using the SUMMER package Lunch Break Lecture 7A: SAE (Wakefield). More complex modeling and BRFSS example Lecture 7B: SAE in R (Li). BRFSS example. More on the SUMMER package (simple binary outcome, no time) Coffee Break Lecture 8A: Advanced SAE (Wakefield). Space-time modeling illustrated with Kenya U5MR example. Continuous spatial models and estimation at the pixel level Lecture 8B: Advanced SAE in R (Li) U5MR SUMMER package example. 6 / 6